[CLS] Is Not Enough: Multi-Label Recognition via Patch-Level Inference and Adaptive Aggregation

Researchers introduced PIAA, a training-free patch-level framework for multi-label image recognition.

PIAA moves beyond the single global [CLS] token by inferring on image patches and adaptively aggregating scores.

The method improves patch discrimination and reduces the vision-language gap with minimal extra compute.

It delivers strong benchmark gains, including more than a 6% mAP boost on NUS-WIDE.